The four tools have been deployed in LTU design studios, most extensively in the Comprehensive Design capstone where students must integrate structural, environmental, and spatial systems at the schematic phase — the moment when massing decisions carry the greatest consequences. Each tool was built for a specific pedagogical purpose and a specific student population: graduates entering regional practice, where detailed simulation software is rarely available at the stage when it would do the most good. The scholarly argument running through this work concerns the difference between using AI tools and making them. Specifying what a correct calculation should do — precisely enough for a code generator to implement it — forces tacit professional knowledge into the open. An architect who has reviewed structural drawings recognizes immediately when a ratio is wrong; toolmaking is what converts that intuition into something communicable. That externalization is where the learning happens, for students and for the instructor building the tool. A parallel project applied the same approach to faculty grading. A custom analytics dashboard, built with Claude Code, helped four faculty assess 53 students across three work cycles in the Comprehensive Design studio, surfacing scoring divergence between reviewers and tracking calibration over the semester. The class average climbed from 78.6% to 87.1% across the three cycles — whether from improving student work, better faculty calibration, or both, the dashboard made the trajectory visible. This body of work has generated a journal paper submitted to Base Diseño e Innovación, a Field Note contribution accepted to ACADIA 2026, and a workshop accepted at EDRA 57 — a 90-minute format where faculty and graduate students with no programming background prototype working domain-specific tools using AI assistance. The workshop format is itself an argument: toolmaking is a learnable practice, not a technical prerequisite. (pending peer review)
My scholarship investigates AI-assisted toolmaking as a distinct form of architectural inquiry and pedagogical practice. Between 2024 and 2026, I developed four production-grade, browser-based design tools — a parametric mass timber structural designer, an acoustic ray tracing visualizer, an HVAC load calculator, and an AI-powered explorer of Christopher Alexander’s A Pattern Language — each built through iterative natural-language prompting without traditional programming. The broader scholarly argument is that the act of making computational tools, rather than consuming them, develops algorithmic literacy through specification: when an architect must articulate what a correct calculation should do precisely enough for a code generator to implement it, tacit professional knowledge becomes explicit, testable, and communicable.
